Effects of Hilling Application to Mitigate Damage from Soybean Gall Midge

Tyler Wiederich

University of Nebraska-Lincoln (STAT 930)

5/18/23

Soybean Background

Soybean Background

  • Soybean is an essential crop in the United States
  • Used for food and oil
    • Primarily for feeding livestock and producing cooking oil

Soybean Gall Midge

A new pest

  • In 2011, there were reports of an unidentified orange larvae on the stems of damaged soybeans after a hailstorm
  • Entomologists received reports of larvae infestations in 2018 for an unknown species causing significant damage

A new pest

  • Initial surveys indicated the infestation was in Minnesota, Iowa, South Dakota, and Nebraska
  • As of 2020, the insect was identified in 114 counties, now including Missouri

A new pest

  • Initially found on damaged plants at the end of the growing season, causing little economic concern
  • In 2018, an infestation warranted further concerns when a sample of damaged plants from earlier in the growing season had no other detectable plant diseases
  • Insects were identified as a new species in the Resseliella genus
  • Females lay eggs in damaged parts of soybean stems (including naturally occurring fissures) and larvae eat away at the stem

Possible solutions

  • At this time, there are no guaranteed methods for preventing damage due to soybean gall midge

  • One possible method, the subject of today’s talk, is to cover the stems of soybeans with dirt

    • A process known as hilling

Two Studies on the Effects of Hilling

Hilling Study

Does hilling at a particular stage in the soybean growth cycle help to decrease the number of soybean gall midge (SGM) larvae?

  • In 2022, a study was conducted as a randomized complete block design (RCBD) and measured the count of SGM larvae.
    • An infested field had four rows where each row was used as a block
    • Within each field row, four treatments were randomly applied to a section
      • No hilling (control)
      • Hilling at V2, V5, and R2 soybean growth stages
    • Measurements were taken at approximately two-week intervals

Hilling Study: Fitted Model

\[\begin{equation} \eta_{ijk}=\eta + \tau_i + S_j + (\tau S)_{ij} + (B \tau)_{ik} + S(B\tau)_{ijk} \end{equation}\]

Where

  • \(\eta\) is the intercept

  • \(\tau_i\) is the \(i^{th}\) hilling date

  • \(S_j\) is the \(j^{th}\) sample date

  • \((\tau S)_{ij}\) is the interaction between the \(i^{th}\) hilling date and the \(j^{th}\) sample date

  • \((B \tau)_{ik}\sim N(0, \sigma^2_{A})\) is the random interaction effect between the \(i^{th}\) treatment and the \(k^{th}\) field row

  • \(S(B\tau)_{ijk}\sim N(0, \sigma^2_{B})\) is the random effect of the \(i^{th}\) hilling date, \(j^{th}\) sample date, and the \(k^{th}\) row

  • Distribution: \(y_{ijk}|(B \tau)_{ij}, S(B\tau)_{ijk}\sim Poisson(\lambda_{ijk})\), where \(y_{ijk}\) is the count of SGM larvae

  • Link function: \(\eta_{ijk}=\log(\lambda_{ijk})\)

Hilling Study: Results

  • Some evidence of interaction between sample date and timing for application of hilling
  • Only a few differences detected
    • July 5, differences between control and V2 (Treatments 1 and 2, p-value = .0203) and between V2 and R2 (Treatments 2 and 4, p-value = .0351)
  • Possible evidence that hilling at the V2 stage help reduce SGM larvae counts?

Unhilling Study

Two primary research questions:

  1. How does unhilling affect total SGM larvae counts? / Does having a preventative measure earlier in the growing season decrease SGM larvae counts?

  2. How does unhilling affect soybean growth/yield?

  • Study design is similar to the hilling study (RCBD)

    • Field organized into two rows and two columns, with each section acting as a block

    • Seven treatments of unhilling dates approximately two weeks apart (one left unhilled at the start of the study as a control)

    • For (1), larvae counts were taken approximately every two weeks

    • For (2), growth/yield metrics were taken at the end of the growing season

Unhilling Study: Fitted Model (1)

\[\begin{equation} \eta_{ijk}=\eta + \tau_i + S_j + (\tau S)_{ij} + S(B\tau)_{ijk} \end{equation}\]
  • Terms are defined the same as the hilling study
  • \(y_{ijk}| S(B\tau)_{ijk}\sim Poisson(\lambda_{ijk})\)
  • \(\eta_{ijk}=\log(\lambda_{ijk})\)

Unhilling Study: Results (1)

Differences in unhilling dates for sample date C

Difference Treatments Effect P-value
Control and July 15 1, 4 111 0.0515*
Control and August 1 1, 5 119 0.0202
Control and August 15 1, 6 118.5 0.0215
Control and August 31 1, 7 114.25 0.0356
June 16 and July 15 2, 4 147.24 0.0004
June 16 and August 1 2, 5 155.25 0.0001
June 16 and August 15 2, 6 154.75 0.0001
June 16 and August 31 2, 7 150.5 0.0002

Differences in unhilling dates for sample date D

Unhilling Differences Treatments Effect P-value
Control and July 15 1 - 4 -135.75 0.0022
Control and August 1 1 - 5 105 0.0973*
Control and August 15 1 - 6 128.5 0.006
Control and August 31 1 - 7 142.25 0.0009
June 16 and August 1 2 - 5 141.75 0.0009
June 16 and August 15 2 - 6 111.5 0.0487
June 16 and August 31 2 - 7 125.25 0.0092

Differences marked with an asterisk (*) are considered marginally significant. There were no other significant simple effect differences for each sample date.

Unhilling Study: Results (1)

Larger treatment numbers denote a later date for unhilling. Trends indicate that unhilling earlier in the season have larger counts of SGM larvae.

Unhilling Study: Fitted Model (2)

The full model is presented for the five collected responses. These are count of soybean nodes, pods, seeds, and plant height and seed weight.

\[\begin{equation} \eta_{ijk}=\eta + B_i + \tau_j + (B\tau)_{ij} + \epsilon_{ijk} \end{equation}\]

Where

  • \(\eta\) is the intercept

  • \(B_i\sim N(0, \sigma^2_B)\) is the effect of the \(i^{th}\) field section (block)

  • \(\tau_j\) is the effect of the \(j^{th}\) unhilling date

  • \((B\tau)_{ij}\sim N(0, \sigma^2_{B\tau})\) is the interaction effect of the \(i^{th}\) field section and the \(j^{th}\) unhilling date

  • \(\epsilon_{ijk}\sim N(0, \sigma^2_e)\) is the random error of the \(i^{th}\) field section, \(j^{th}\) unhilling date, and the \(k^{th}\) plant

The model is then fit the these specifications

Response Distribution Link Function Changes from full model
Count of nodes \(y|B\sim Poisson(\lambda)\) \(\eta_{i}=\log(\lambda_i)\) Removal of \(B_i\) and \(\epsilon_{ijk}\)
Soybean height \(y|B\sim Normal(\mu, \sigma^2)\) \(\eta_{i}=\mu_i\) Use of CS covariance structure instead of \(B_i\) term
Count of pods \(y|B\sim Negbin(\lambda)\) \(\eta_{i}=\log(\lambda_i)\) Removal of \(B_i\) term and \(\epsilon_{ijk}\); KR2
Seed weight \(y|B\sim Normal(\mu, \sigma^2)\) \(\eta_{i}=\mu_i\) No adjustments
Count of seeds \(y|B\sim Negbin(\lambda)\) \(\eta_{i}=\log(\lambda_i)\) Removal of \(B_i\) term and \(\epsilon_{ijk}\); KR2

Unhilling Study: Results (2) - Soybean Height

Unhilling Date Estimate
August 15th 77.375 A
A
August 1st 75.35 A
A
July 15th 73.6 A
A
July 1st 70.4 A
A
August 31st 64.025 B A
B
June 16th 50.975 B
Unhilled (control) 32.85 C

In general, unhilling earlier in the growing season decreases soybean height

Unhilling Study: Results (2) - Soybean Height

Unhilling Study: Results (2) - Soybean Seed Count

Unhilling date Estimate
August 15th 118.96 A
A
August 1st 100.81 A
A
July 15th 96.87 A
A
August 31st 94.26 B A
B A
July 1st 77.80 B A
B A
June 16th 8.05 B A
B
Unhilled (control) 5.12 B

Unhilling Study: Results (2) - Soybean Seed Count

Similar to soybean height, unhilling earlier in the season generally decreased soybean seed counts

Discussion

Discussion

  • Infestation has possibility to cause ecological and economical harm

  • Protecting soybean stems via hilling earlier in the season decreased SGM larvae counts and improved soybean growth/yield metrics when there was an active SGM infestation

  • No method completely prevented SGM larvae from getting into the stems

STAT 930 Reflection

STAT 930 Reflection

  • Good clients overall this semester
  • No follow-ups with clients

  • Some issues with communications

    • A couple of clients either “ghosted” me or took a long time to reply
  • Client projects

    1. Measurements on different types of asphalt
    • One factor experiments where there were sometimes one experimental unit
    1. Survey data on crop fields with a particular disease
    • Lots of graphs and tables to answer “what are some commonalities in the data?”
    • Directed to NEAR center for future surveys
    • Client not sure of research question
    1. Bike trails and improvements on quality of life metrics
    • Small sample size and nothing was significant
    1. Soybean pest infestation (project from today’s talk)
    • Many GLMMs
    • Models did not converge with “full models.”
    1. Student performance and behavior surveys
    • Directed to NEAR center

Questions?